Danzek SGE Journal
2012 March 12
Why are resource-reservations not working properly? Discussion lists (see below) suggest that there is a bug/feature. The default value of default_duration is INFINITY and this causes resource-reservations to fail sometimes. Change to queue length for example. . .
qconf -ssconf . . default_duration 168:00:00 .. . .and problem goes away?
Diagnostics:
tail -f /$SGE_ROOT/default/common/schedule | grep RESERVor
tail -f /$SGE_ROOT/default/common/schedule | grep <JOB_NUM_WITH_RR> # ...that's the id/number of the job on which a r-r was set...
http://permalink.gmane.org/gmane.comp.clustering.gridengine.users/19485:
Re: 6.2u4: resource reservation not working for some jobs Hi Richard, rems0 schrieb: > Hi list, hi reuti, > > resource reservation is still not working for me. > In the scheduler configuration (qconf -ssconf) > max_reservation is set to 20. > default_duration is set to 8760:00:00. are you monitoring your scheduler? ------------------------- >8 ------------------------- qconf -ssconf | grep params params MONITOR=1 ------------------------- 8< ------------------------- The location of the schedule file is: <sge_root>/<cell>/common/schedule Do you find any RESERVING statements in the log file? > As mentioned in the bug above, I changed default_duration in the > scheduler configuration from INFINITY to 8760 hours (1 year). > > Should this be enough or should I also change s_rt and/or h_rt in the > queue definitions? These are both also set to INFINITY. This should be enough.
2012 March 12
Problems with one or two MACE users hogging the IB-connected queues. This is done by exploiting the urgency policy preceding this date which was:
Weight Deadline : zero Weight Waiting Time : zerowith
bash> qconf -sc #name shortcut type relop requestable consumable default urgency #------------------------------------------------------------------------------------------ . . . . . . . . . . . . . . . . slots s INT <= YES YES 1 1000 . . . . . . . . . . . . . . . .
Urgency Policy Change:
Weight Waiting Time : 0.1
Resource Quota Changes:
-- added 256 slot global limit for each user